Decoding fairness: A reinforcement learning perspective.

Area of Science:

• Behavioral Economics
• Computational Neuroscience
• Game Theory

Background:

• Behavioral experiments on the ultimatum game (UG) show humans prefer fairness, contradicting orthodox economic predictions.
• Existing explanations often rely on external (exogenous) factors within imitation learning frameworks.

Purpose of the Study:

• To investigate the emergence of fairness in the ultimatum game using a reinforcement learning (RL) paradigm.
• To determine if endogenous incentives alone can explain fairness without external factors.

Main Methods:

• Applied Q learning to the ultimatum game, assigning two Q tables per player for proposer and responder roles.
• Analyzed a two-player scenario, extending to latticed populations with various role assignments (random, fixed, rotating).

Main Results:

• Fairness emerged prominently in the UG when players valued both experience and future rewards.
• The probability of successful deals increased with higher offers, aligning with empirical observations.
• The system exhibited two phases, stabilizing into either fair or rational strategies, robust across different conditions.

Conclusions:

• Endogenous incentives within a reinforcement learning framework are sufficient to explain the emergence of fairness in the ultimatum game.
• Exogenous factors are not necessary to account for observed fair behavior in economic decision-making.